Design And Implementation Of Fine Classification Methods For Multi-load Users At Different Time Scales

With the rapid development of Chinese smart grid,the interaction between power enterprises and load users has grown rapidly,which has led to the accumulation of a large number of user-side load data.Considering the tense power system environment,China gradually promotes the shift of the focus of the work of power enterprises,from the single demand side management to the user-side resources actively participate in the power system supply and demand balance process transfer.At this stage,power system load data has the characteristics of high density and diversity,the complex load environment makes the user-side data characteristics mining insufficient,the deep mining user's power behavior is the basis to support the user-side big data value mining,in order to improve the utilization rate of electric energy,It is very important to further excavate the operating value of different types of user load.In the process of deeply digging the value of load characteristics,the classification of power load users is a key step.Fine classification of load users is also one of the key research topics of smart grids.Traditional user load classification has been inappropriate for multiple markets.It is more suitable for the current user market classification method.Firstly,data pre-processing was performed on the multi-load historical operation data of Jilin Province to remove the influence of abnormal data,and “horizontal-vertical” outlier processing was used to fill the missing data.Then performing the first cluster analysis on the preprocessed data,it would be to optimize the initial clustering center,set the optimal K value and improve the K-means clustering algorithm to achieve better clustering results.The load characteristics of users are grouped into the same category.According to the daily load curve of users in the same category after clustering,it can be seen that some users in the same category have a large difference between their electricity consumption behavior and the load characteristics of this category.In order to provide more reliable and finer user classification for power companies,this paper proposes to perform secondary clustering on load users to obtain more refined user classification,extract different types of load characteristic curves after fine division,and establish a load classification model for power users in Jilin Province.After the classification model is established,the historical load data of Jilin Province power grid is trained and tested.Experiments are performed using common classification algorithms such as SVM,KNN,and decision tree.The parameters in each classification algorithm are adjusted to select a classifier suitable for the data set in this paper..The SVM is selected as a classifier for fine classification of load users by the evaluation index,which can realize a more applicable power user load classification model.It is verified by practice that the results obtained by using the power load user classification model of Jilin Province proposed in this paper have a high accuracy rate for user classification.According to the power load user classification model proposed in this paper,power companies can classify users,predict user categories,and provide different power supply needs.In general,the secondary clustering proposed in this paper achieves fine division of load users,and the use of SVM to realize user category judgment has certain significance,providing a more scientific basis for future grid planning and decision-making work.